The Research Of Moving Object Detection And Shadow Elimination Based On Video Imagery

As a basis of moving target video analysis technology,video moving target detection and shadow removal has become a key research project of intelligent monitoring system,behavior recognition and understanding systems,and human-computer interaction systems.Its task is to draw out of our concern moving targets as undistorted as we can from a complex background scene,to lay a foundation for the subsequent conduct of the target tracking,behavior recognition and scene understanding,etc.However,because the disturbance of complicated video scene is so frequent that the timeliness and accuracy of foreground extraction and shadow removel can't be satisfied.Therefore,the paper will advance an improved foreground extraction and shawdoe removel algorithm.The major studies are as follows:Firstly,concerning the foreground detection,this paper combines the most real-time frame difference method with the highest detection accuracy Gaussian mixture background model to advance an improved Gaussian mixture background modeling algorithm.First,the improved algorithm uses a new filtering method to deal with image pre-processing,and can remove the most noise of image without blurring marginal information,according to the LBP value of the pixel point of video.Second,it values new method to improve the initial effect by extracting the pixel value of the first initial 1,11,21,…,(10 N-9)frame image as the initial mean values to be gived N Gaussian distribution to improve the effect of background initial.Final,using frame difference method extracts the zone of background modeling.By superimposing the results of the adjacent seven frames of the differenct result between the first and forth frame with the different result of the forth and seventh,extracting enough zone can get the effect that reduces the time of background modeling.This experiment shows that the algorithm not only reach the effect of Gaussian mixture background modeling,while also shorten the average processing time per frame to improve the operating speed of the system.Secondly,For shadow removal module,in the case of a comprehensive comparative analysis of the advantages and disadvantages of various shadow removal algorithm,according to the scene background texture feature clear whether or not,this paper proposes two shadow removal algorithms.First,having clear background texture feature,by calculating LBP texture feature of checked pixels of moving foreground,and comparing with the LBP texture feature of relevent background video pixels.If the values is equal,the point can be judged as shadow point.Second,for the blurring background texture feature scenes,by drawing separately margin of extracted foreground and live foreground images,and making difference between the two different values,it can gets the marginal testing videos of foreground target that is removed shadow.The two algorithms have the advantages that calculation amount is small,and the testing effect is accurate,and the effectiveness of the algorithm can be also demonstrated by respectively testing and removing foreground shadow.Finally,the lastest part summarizes the difficulties of this algorithm and prospects the insufficiency of different algorithms.